I am, again, still wondering what sort of new insight that these papers brings to view. On the corollary I also read some of Cabot's work, and they're all basically forms of a similar problem being iterated in different fashion. Pick the flavor that makes you happy.
Likewise, I can appreciate the utility of some of Hirato's work in an era where visualizing FFT's wasn't really a thing like it is today and THD+N measurements meant integrating what's left when you null the stimulus with the response. If those are the measurements you're doing nowadays and that THD+N isn't TINY (who cares what the composition looks like when it's sitting in the weeds of a few PPM or so), then you get what you asked for.
It's like someone's going to town on a strawman, at least in terms of the level of discussion on this thread.
So, in all honesty, can someone inform this amateur (that's totally fine, we all have our domains of expertise), what's the advantages of these waveforms are that makes them somehow more useful? And not by saying "try it yourself" because that's a cop out for not being able to elucidate (and perhaps understand) the fundamentals oneself. Insert Feynman quote here about giving a freshman level lecture on a topic to really test your understanding.
Likewise, I can appreciate the utility of some of Hirato's work in an era where visualizing FFT's wasn't really a thing like it is today and THD+N measurements meant integrating what's left when you null the stimulus with the response. If those are the measurements you're doing nowadays and that THD+N isn't TINY (who cares what the composition looks like when it's sitting in the weeds of a few PPM or so), then you get what you asked for.
It's like someone's going to town on a strawman, at least in terms of the level of discussion on this thread.
So, in all honesty, can someone inform this amateur (that's totally fine, we all have our domains of expertise), what's the advantages of these waveforms are that makes them somehow more useful? And not by saying "try it yourself" because that's a cop out for not being able to elucidate (and perhaps understand) the fundamentals oneself. Insert Feynman quote here about giving a freshman level lecture on a topic to really test your understanding.
I think there is no need for such a high band as we are not in concern here with IM produced by high harmonics and original signal like in the measured gear. The difference is *the* distortion by definition, of very low level. And, really, I'm not afraid if some contents of this distortion are <-90dB and > 15Khz.
The only thing that is a little obscure for me is that phase rotation across the measured gear has a big impact. Not sure our ears are so sensible to this, apart for localisation.
Of course, due to the delais introduced by the slow speed of the sound in air, and their lack of linearity, Loudspeakers will not be simple to measure by this method.
If I can't explain it to a freshman, I have not understood it myself - I have quoted this paraphrase many times myself and it's a good one. My favourite though is Niels Bohr "Don't explain yourself more clearly than you are able to think." That's a tough ask for many and let's say all of us?
Why do you even think 768 kHz sampling is desirable? It's well past innocuous overkill at that point. The converters perform worse at those crazy high sample rates. Clearly not the BEST DIGITAL.
I already have a dual 512K-24bit measurement system in my Stanford Research SR-1, and I don't think that it is good enough to do this test correctly. I could try it, I guess, the last time I tried it was 44 years ago with 50K-10bits, and I could measure NOTHING with my class A preamp circuits. It would be better with the new system today, but I still don't think that it will have enough resolution to give me usable results. Special test waveforms seem more appropriate, and may bring out the subtle distortions that make all the difference in audio electronics sound quality. FM distortion would be a paramount consideration, but it might be easier and more definitive to measure it with an FM test apparatus, rather than an FFT.
@John, thanks for posting the info including your work and Hirata's data.
@ Scott, thanks for Sisyphus, which lead me here:
Local boy made good: LINK
I've always loved this song. I can't discern how it would sound in a real stereo system as I'ts on YouTube. It's clean sounding but had low mid bloat that I'm sensitive too. I wish I could get a clean data stream into my "system" and not just on the 'puter.
Surely there has to be a simple method to do this, then what? Go into my pre amp using the AUX or
one of the Tape inputs?
Another classic here Miles: LINK
and this too: LINK
In all of these newer releases of Take Five they've removed from the original recording and pressing the counting from the bassist in the quiet passages of the drum solo. Any one with an original pressing will hear this when you turn up the volume...
Cheers,
@ Scott, thanks for Sisyphus, which lead me here:
Local boy made good: LINK
I've always loved this song. I can't discern how it would sound in a real stereo system as I'ts on YouTube. It's clean sounding but had low mid bloat that I'm sensitive too. I wish I could get a clean data stream into my "system" and not just on the 'puter.
Surely there has to be a simple method to do this, then what? Go into my pre amp using the AUX or
one of the Tape inputs?
Another classic here Miles: LINK
and this too: LINK
In all of these newer releases of Take Five they've removed from the original recording and pressing the counting from the bassist in the quiet passages of the drum solo. Any one with an original pressing will hear this when you turn up the volume...
Cheers,
Last edited:
The best ADC's for linearity for audio run at 192 KHz. The distortion products are close to -130 dB. Audio Diffmaker Audio DiffMaker can run at 24/192. With a digital source there will be very little "deskewing" to deal with.
You could do the Keith Johnson test and simply record the signal at the input on one channel, pass it through something and record the output on the other channel. gain trim and invert one and sum. What you hear will be the differences. He uses that o verify digital eq for problems. What you are describing should fall into that same basket.
This is nothing new and has been discussed here before. I'm not sure how you would specifically look at FM issues.
You could do the Keith Johnson test and simply record the signal at the input on one channel, pass it through something and record the output on the other channel. gain trim and invert one and sum. What you hear will be the differences. He uses that o verify digital eq for problems. What you are describing should fall into that same basket.
This is nothing new and has been discussed here before. I'm not sure how you would specifically look at FM issues.
Yes. And what does a harmonic distortion analyzer do with such differences between polarities?
....and summed to a much lower level on OP. But does it Sound different to have no differences in pol distortion compared to one which shows different pol distortions?
THx-RNMarsh
Last edited:
?? What is 'pol distortions'?
If the waveform at the OP (I assume that means output) is asymmetric you have even order harmonics. If the waveshape is symmetric you have odds. Surely evens or odds sound different, if they are large enough?
I really would think these basic issues would be understood before moving to Hirata stuff?
Jan
Edit: to amplify, suppose you have a signal that shows some slew rate limiting at one polarity only. That would make the waveform asymmetric and if you do a THD measurement on that signal you'd see even harmonics. But from looking at those harmonics you probably can't say what the cause is, except that it is from an asymmetric wave.
That is of course why we say that push-pull cancels even harmonics - it doesn't really do that but the push-pull nature distorts both signal polarities equally thus making a symmetric albeit distorted wave which then shows odds predominantly. So by looking at the Fourier spectrum you can have a pretty good idea whether you are looking at a single-ended or push-pull amp.
All basic stuff of course but sometimes we get ahead of ourselves.
Last edited:
Yes and if you think about it a perfectly square law circuit that changes polarity in the third quadrant ("S") shape has no second harmonics at all.
This feels like a JC question... a red herring. Different goals will have different differences between your set of questions. You can't pretend you dont know about things like pnp/npn transconductance. Maybe you mean to ask the significance of the differences?
Electronics are never perfectly symmetrical, but can accepting some asymetry to encourage uneven benefits between rails be a good thing? That to me is a more ponderous question since we know they are not equal.
... and that was exactly a point (among others) of Ian Hegglun's 'square law class A' amps he wrote about in Linear Audio.
If you stay in 'Fourier denial' you miss so much insight that really is enjoyable!
Jan
Diff test is really revealing and has tremendous resolution when a) using analog diff and gain before the ADC (will need extensive trimming to null most of the linear differences but for simple DUTs it is doable without major efforts), and b) using low jitter synchronous averaging (floating point numbers, of course) with sample-sync'd AD and DA to lower any noise not strongly correlated to the signal (30dB and more are feasible).
Nice thing is you can use an arbitrary signal for the test and actually listen to the residual.
Even when the trimming and diff process is done in the digital domain (like DiffMaker does) there is tons of resolution, the key element is the sync'd time-domain averaging (before the diff process, in this case). The drawback of heavy averaging is that is blind for spurious loosely correlated errors signals.
Nice thing is you can use an arbitrary signal for the test and actually listen to the residual.
Even when the trimming and diff process is done in the digital domain (like DiffMaker does) there is tons of resolution, the key element is the sync'd time-domain averaging (before the diff process, in this case). The drawback of heavy averaging is that is blind for spurious loosely correlated errors signals.
Last edited:
It seems the test JC put up are more useful in helping to identify where a circuit issue is coming from than would looking at FFT of harmonics. As would seeing asym slew rates.
Each different test has its benefits and insights into circuit behaviour. All show versions or portions of the same over-all performance. They help to direct the design effort to the cause of that 3rd or 2nd etc shown in your FFT. This is not the same as just knowing the distortion by itself. A diff comparator doesnt tell you anything about the circuit operation, either.
And, it will help you to know better and identify what the circuit will sound like as another descriptor.
Just sayin.
THx-RNMarsh
Each different test has its benefits and insights into circuit behaviour. All show versions or portions of the same over-all performance. They help to direct the design effort to the cause of that 3rd or 2nd etc shown in your FFT. This is not the same as just knowing the distortion by itself. A diff comparator doesnt tell you anything about the circuit operation, either.
And, it will help you to know better and identify what the circuit will sound like as another descriptor.
Just sayin.
THx-RNMarsh
Last edited:
Dick, that maybe true, but that's all we hear, right, 'it seems', 'maybe'.
JC just throws up a paper, makes a statement to the effect that we all are dummies and have no clue, and that's it. We don't even know if he understands the stuff he puts up himself.
Why doesn't he then, or you for that matter, explain what the test does, what it shows, where it comes from, what it implies for the circuit.
Just throwing up some random stuff is very cheap and unsatisfying. Or maybe we should accept that that is all we are getting in The Lounge. Just sayin.
Jan
JC just throws up a paper, makes a statement to the effect that we all are dummies and have no clue, and that's it. We don't even know if he understands the stuff he puts up himself.
Why doesn't he then, or you for that matter, explain what the test does, what it shows, where it comes from, what it implies for the circuit.
Just throwing up some random stuff is very cheap and unsatisfying. Or maybe we should accept that that is all we are getting in The Lounge. Just sayin.
Jan
- Status
- Not open for further replies.